Menu
Enterprise AI design patterns for production systems - RAG, Multi-Agent, AI Gateway, LLMOps, and more
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Apply official FrankX brand standards to all artifacts, ensuring visual consistency across content, products, and communications. Use this skill for any FrankX-related content creation, design work, or communication.
Frank's personalized career and professional mastery coaching
Guided process to create your personalized Soulbook - a comprehensive life transformation system
Creates expert positioning content, social media posts, and marketing materials for Frank's personal brand and AI coaching business with soul-aligned messaging
Master-level visionary book writing system with research-driven methodology, author voice modeling, and iterative refinement across 9 quality dimensions
Expert Suno AI prompt engineering for cinematic, transformative music creation. Use this skill when creating Suno prompts for Vibe OS sessions, meditation tracks, or any AI-generated music that needs professional quality and emotional resonance.
| name | AI Architecture Patterns |
| description | Enterprise AI design patterns for production systems - RAG, Multi-Agent, AI Gateway, LLMOps, and more |
| version | 1.0.0 |
| category | technical |
| command | /ai-architecture-patterns |
| triggers | ["ai pattern","architecture pattern","rag pattern","agent pattern","ai gateway"] |
| type | reference |
Provide expert guidance on selecting and implementing enterprise AI architecture patterns for production systems. This skill contains battle-tested patterns from real-world deployments and the AI Architect Academy.
Problem: Multiple AI services with inconsistent interfaces, no centralized security, and limited observability create management complexity and security risks.
Solution: Deploy a centralized AI gateway that provides unified authentication, rate limiting, request/response logging, and model routing for all AI services.
Key Components:
When to Use:
When NOT to Use:
Problem: LLMs hallucinate and lack access to enterprise-specific knowledge, making them unreliable for business-critical applications.
Solution: Implement a RAG pipeline with document ingestion, chunking, embedding, vector storage, retrieval, and augmented generation with source citations.
Key Components:
When to Use:
When NOT to Use:
Implementation Tips:
Problem: Complex tasks require multiple specialized capabilities that exceed what a single LLM prompt can handle reliably.
Solution: Decompose complex workflows into specialized agents with an orchestrator that coordinates task distribution, handoffs, and result aggregation.
Key Components:
When to Use:
When NOT to Use:
Frameworks:
Problem: N agents x M tools = N*M integrations. Each AI agent needs custom code to integrate with each tool.
Solution: Implement MCP (Model Context Protocol) servers that provide standardized interfaces for tools, resources, and prompts.
Key Components:
When to Use:
Implementation:
import { Server } from '@modelcontextprotocol/sdk/server';
const server = new Server({
name: 'my-mcp-server',
version: '1.0.0'
});
server.tool('search', {
description: 'Search documents',
inputSchema: {
type: 'object',
properties: {
query: { type: 'string' }
}
},
handler: async ({ query }) => {
// Implementation
}
});
Problem: LLM applications lack mature DevOps practices, leading to unpredictable quality and difficult rollbacks.
Solution: Implement prompt versioning, automated evaluation, staged deployments, and continuous monitoring.
Key Components:
When to Use:
Evaluation Metrics:
Problem: Many vector database options with different tradeoffs. Wrong choice leads to expensive migrations.
Solution: Structured decision framework evaluating scale, features, operations, and cost.
Selection Matrix:
| Scale | Recommendation |
|---|---|
| <1M vectors | pgvector (simple), Chroma (prototyping) |
| 1-100M vectors | Weaviate, Qdrant (self-hosted) |
| 100M+ vectors | Pinecone, Milvus (managed) |
Key Considerations:
Problem: Scattered AI initiatives across organization lead to duplicated effort, inconsistent quality, and security gaps.
Solution: Establish centralized governance with standardized patterns, reusable components, and shared infrastructure.
Key Components:
Governance Areas:
Problem: AI introduces new security vectors: prompt injection, data leakage, model manipulation.
Solution: Implement AI-specific security controls including guardrails, PII handling, and audit logging.
Key Controls:
Guardrails Implementation:
from guardrails import Guard
guard = Guard.from_pydantic(output_class=SafeResponse)
response = guard(
llm.invoke,
prompt=user_input,
on_fail="reask"
)
START: What type of AI system?
│
├── Document/Knowledge Q&A
│ └── → RAG Production Pattern
│ ├── Need multiple models? → + AI Gateway
│ └── Sensitive data? → + Security & Governance
│
├── Autonomous Agents
│ └── → Multi-Agent Orchestration
│ ├── Many tools? → + MCP Servers
│ └── Production deployment? → + LLMOps
│
├── Enterprise AI Platform
│ └── → AI Gateway + AI CoE Framework
│ ├── Cost concerns? → + Cost Optimization
│ └── Compliance? → + Security & Governance
│
└── Content Generation
└── → AI Gateway + LLMOps
└── Quality critical? → + Evaluation Pipeline
| Use Case | Primary | Secondary | Tertiary |
|---|---|---|---|
| Customer Support Bot | RAG | AI Gateway | Security |
| Code Assistant | Multi-Agent | MCP Servers | LLMOps |
| Document Intelligence | RAG | Vector DB | AI Gateway |
| Enterprise AI Platform | AI Gateway | AI CoE | Security |
| Research Assistant | RAG | Multi-Agent | LLMOps |
mcp-architecture - MCP server developmentclaude-sdk - Agent development with Claudelanggraph-patterns - Graph-based agent workflowsoci-services-expert - Oracle Cloud guidancePart of the AI Architect Academy by FrankX.AI